When Zillow's latest rent report estimated San Francisco's median rent at $4,225 last month, the news put the fear of God and math in us all, garnering headlines here and, last week, over at SFGate's On the Block blog. Zillow's number has lately met with some backlash, however. SocketSite declared "San Francisco's Median Rent Is NOT $4,225. Really, It's Not" in one story, which was then amplified by SFist's "Stop Saying That SF's Median Rent Is $4225." According to both sites, Zillow's number can't be accurate because it's "based on estimated, versus actual, rents" and "doesn't include … apartments!" in the words of SocketSite. To which we naturally said, WTF? We called Skylar Olsen, senior economist for Zillow, to find out where the number comes from.
Unlike the monthly rent reports released by rental website Zumper, Zillow's rent index doesn't simply gather up one month's rental listings and find a median. Though figures like Zumper's provide an accurate picture of what advertised apartments are renting for each month, they're pretty bad at measuring a change in prices meaningfully over time. That's because the one-bedrooms renting this month aren't necessarily the same ones that were advertised last month. So if June's median is up $40 from May's, it's impossible draw any real conclusion from that. It would not be accurate to say that the median apartment has gotten $40 more expensive, because the sample we're looking at has changed.
Zillow gets around this, Olsen explains, by using an elaborate algorithm that analyzes real price data in order to assign an overall value to San Francisco's housing stock at any given time. The point of doing this, she says, is to standardize this constantly changing rental market data in a way that produces an "apples-to-apples comparison" over time and across different metro areas.
Here's how it works:
Zillow mines the data from its rental listings to understand the relationship between advertised rental price and bedroom count, square footage, and location (plus a long list of other factors, but those are the important ones). With every new rental listing, the algorithm learns more about how much people are likely to be willing to pay for, say, 100 extra square feet in Russian Hill versus Western Addition.
The algorithm's expertise at matching rental attributes with dollar amounts means that Zillow can look at the whole of a city's housing stock, not just what's on the market at any given time, and add up its market value. The point of doing this is to show how prices are changing over time. If you're analyzing the value of all known housing stock each month, comparing price changes month to month suddenly becomes possible. So Zillow can say, in real terms, that rents in San Francisco rose 15.5 percent from the first quarter of 2014 to the first quarter of 2015. Not only that, but a comparison between, say, Oakland's rents and San Francisco's becomes feasible.
And here's where we get to the "whaaa?" that's fueling the backlash. In effort to work the numbers into a place where they can accurately reflect price differences, Zillow has devised a mathematical landscape that doesn't look quite like the San Francisco rental market we know and gripe about. It's not an absolute median rent (which would require a survey of what everyone in the city is paying, and which no one has), and it's not a median asking rent, which is Zumper's beat. Instead, Zillow's $4,225 describes the median rent in an artificial scenario in which San Francisco's entire housing stock is listed on the rental market all at once and can magically command market-rate prices. In other words, it's a landlord's fantasy realm.
And now we get to the question of whether the Zillow Rent Index takes real apartments into account. Yes and no. The yes part: Zillow's inventory of apartment listings in San Francisco is the raw data that helps the algorithm learn how much various rental attributes, from size to location to bed and bath count, are worth in dollars. But (here's the no part) when Zillow spits out its median rent, that number is based not on rental housing stock, but on single-family homes, condos, co-ops, and TICs.
Because no one knows how many apartments are available for rent. It doesn't work to simply count up all the apartments listed ever, because there isn't a one-to-one relationship between the number of listings and the number of apartments. Some listings advertise multiple apartments; some apartments have never been listed. Unlike for-sale homes—for which there's a single public record to go with every single home—no public records exist that match up real-life units with leasing information. Because of this black hole in the data, "we shouldn't involve apartments in that overall median," explains Olsen. In other words, why do all that work to standardize apartment market values if you're going to turn around and use it on a by-definition incorrectly tallied number of apartments?
Still, something feels odd about this switcheroo. Just because the exact tally of apartments in San Francisco isn't knowable doesn't make the substitution of for-sale homes in their place suddenly more accurate. It almost seems worse, somehow.
We put this question to Olsen. "It's a difficult problem," she acknowledges. "They end up being substitute goods, in the sense that what an apartment can rent for in one building—its features and whatever else—will affect what a condo unit can rent for next door. They do influence each other. It's a related market."
For an economist's purposes, grafting rental price data onto known condos and co-ops with similar attributes is a logical way around the unknowables in the apartment market. "Nothing out there right now is perfect," says Olsen. "We think our methodology is a useful tool for consumers to think about how rents change over time and to think about rents in different areas. It's not perfect. Is it the best and is it responsible? We think so."
So while the Zillow Rent Index does not present a San Francisco we can exactly recognize, it does tell us some true things, such as the percentage increase in rental prices over time, and how much more expensive (in real terms) renting in San Francisco is than renting in Oakland or San Jose. Our advice: Think of the index as a carefully calibrated but crazy-looking telescope that distorts everything in the viewfinder but miraculously makes the relationships between each object perfectly accurate. Zillow's median rent is both more and less fictional than the medians that only take into account real rents advertised for real apartments.
· San Francisco's Median Rent Climbs to a Whopping $4,225 [Curbed SF]
· San Francisco's Median Rent Hits a Ridiculous $4,225 [SFGate]
· San Francisco's Median Rent Is NOT $4,225. Really, It's Not [SocketSite]
· Stop Saying That SF's Median Rent Is $4225 [SFist]
· Breaking Down SF's Big Annual Rent Increase by Neighborhood [Curbed SF]
· San Francisco Rents Hit an All-Time High, No Sign of Letting Up [Curbed SF]
· Forbes: San Francisco Is the Absolute Worst City for Renters [Curbed SF]